Learn to build effective Agentic AI systems with Andrew Ng
Andrew Ng outlines a course on building agentic AI systems in raw Python, emphasizing core workflows, design patterns, and disciplined evaluation over framework reliance.
Highlights
2 clips · 1 min of 3 min-
- Agentic systems execute multiple LLM-driven steps rather than a single direct response.
- They can utilize tools, reason, and iterate to complete complex tasks.
- Building them using raw Python exposes how each step works without framework abstraction.
-
- Disciplined error analysis via evals is the biggest predictor of team execution success.
- Systematic evaluations let teams efficiently hone in on components that need improvement.
- Data-driven evaluation replaces guessing on how to spend engineering time.
That’s all the highlights.
Questions this video answers
4 · every answer cites the moment it was said01 How do agentic AI systems differ from single LLM calls? 2 sources · 0:33
Agentic systems execute multiple LLM-driven steps, utilizing tools, reasoning, and iteration to complete complex tasks. Instead of relying on a single direct response, they process work iteratively.
02 Why does the course teach agentic workflows using raw Python instead of frameworks? 2 sources · 0:49
Using raw Python avoids hiding implementation details inside a framework, allowing developers to see how each step actually works. This builds a foundational understanding before adopting popular development frameworks.
03 What are the key steps to implementing a complex agentic application? 2 sources · 0:59
Developers should take a complex application and decompose it into a sequence of tasks. They then implement these tasks as an agentic workflow using four key design patterns.
04 How can teams effectively improve their agentic workflows? 2 sources · 1:21
The single biggest predictor of success is driving a disciplined error analysis process using evaluations. Teams should let eval data guide their improvements rather than guessing where to spend their time.
Section-by-section summary
3 parts
Introduction to agentic AI systems
Agentic AI systems have grown rapidly in reality and hype, moving beyond single LLM calls to execute multi-step workflows with tools and reasoning. The course teaches these workflows in raw Python to show the inner workings without hidden framework layers.
Decomposing and designing agentic workflows
Complex applications are broken down into sequences of tasks implemented as agentic workflows using four key design patterns. Understanding these fundamentals allows developers to build robust systems using any framework or none at all.
Driving improvement through systematic evals
Successful agent development relies on a disciplined error analysis process driven by evaluations. Rather than guessing, developers use eval data to pinpoint which components require attention and improvement.
Connects to your library
by shared topics
n next · p previous clip · t highlights/full
< > slower/faster · / search what was said
-
Agentic AI4× · first 0:00
A category of AI projects where systems execute multiple LLM-driven steps with tools, reasoning, and iteration.
-
Design patterns1× · first 1:07
Four key structural patterns used to implement agentic workflows.
-
Error analysis1× · first 1:21
A disciplined process of running evaluations to hone in on components that need improvement.
417 words · about 2 min to read
Introduction to agentic AI systems
Building effective agentic AI systems is one of the most valuable skills in AI today. In this course, I share best practices for doing so and you come away with a systematic understanding of how to build agentic systems. I come with the term agentic AI to describe a growing category of AI projects not
1 Highlight · Agentic AI versus single LLM calls
realizing that marketers would get a hold of that term and slap us a sticker on everything in site. Even though the hype on agentic AI has grown really rapidly, the reality of agentic systems being created and deployed has grown rapidly as well. Unlike a single LM call, they just directly generates a response. Agentic AI systems can execute multiple LM driven steps. These steps can use tools, they can reason, they can iterate over the work to complete complex tasks. In this course you learn how to build agentic workflows using raw Python without hiding all the steps in some framework so that you see
Decomposing and designing agentic workflows
how each step actually works. You learn to take a complex application and decompose it into a sequence of tasks that you can then implement as an agentic workflow. You also learn how to implement the four key design patterns for agentic workflows. Having worked on many agentic systems with many teams, I found that the single biggest predictor for whether the team executes
Driving improvement through systematic evals
2 Highlight · Using evals to guide agentic development
well lies in his ability to drive a disciplined error analysis process. That is to put in place evaluations or evals so that given a complex agented workflow, you can efficiently hone in on what components to focus attention on improving. This way, you aren't just guessing how to spend time productively. you let the eval data guide you. In this course, I'll also show you how to drive this process and this will put you significantly ahead of your game compared to the vast majority of teams building aicles today. This course is taught in a vendor neutral way in raw Python and the emphasis is on teaching
you the core concepts that you can then implement using any of the popular agentic AI development frameworks or using no framework. This long form course is currently available only on the deep learning.ai website. I hope you sign up for this course and when you've completed it, you know how to build aic AI systems which is one of the most in demand skills on the job market today and that will also let you build a lot of cool applications.